21
20
that has merged into it. As the first step of a merge, pull, or
22
21
branch operation we copy history from the source into the destination
25
The copying is done in a slightly complicated order. We don't want to
26
add a revision to the store until everything it refers to is also
27
stored, so that if a revision is present we can totally recreate it.
28
However, we can't know what files are included in a revision until we
29
read its inventory. So we query the inventory store of the source for
30
the ids we need, and then pull those ids and finally actually join
35
import bzrlib.errors as errors
36
from bzrlib.errors import (InstallFailed,
38
from bzrlib.progress import ProgressPhase
25
from __future__ import absolute_import
29
from bzrlib.lazy_import import lazy_import
30
lazy_import(globals(), """
41
from bzrlib.i18n import gettext
39
42
from bzrlib.revision import NULL_REVISION
40
from bzrlib.symbol_versioning import (deprecated_function,
44
43
from bzrlib.trace import mutter
47
from bzrlib.lazy_import import lazy_import
49
# TODO: Avoid repeatedly opening weaves so many times.
51
# XXX: This doesn't handle ghost (not present in branch) revisions at
52
# all yet. I'm not sure they really should be supported.
54
# NOTE: This doesn't copy revisions which may be present but not
55
# merged into the last revision. I'm not sure we want to do that.
57
# - get a list of revisions that need to be pulled in
58
# - for each one, pull in that revision file
59
# and get the inventory, and store the inventory with right
61
# - and get the ancestry, and store that with right parents too
62
# - and keep a note of all file ids and version seen
63
# - then go through all files; for each one get the weave,
64
# and add in all file versions
67
@deprecated_function(zero_eight)
68
def greedy_fetch(to_branch, from_branch, revision=None, pb=None):
69
"""Legacy API, please see branch.fetch(from_branch, last_revision, pb)."""
70
f = Fetcher(to_branch, from_branch, revision, pb)
71
return f.count_copied, f.failed_revisions
76
46
class RepoFetcher(object):
77
47
"""Pull revisions and texts from one repository to another.
80
if set, try to limit to the data this revision references.
83
count_copied -- number of revisions copied
85
This should not be used directory, its essential a object to encapsulate
49
This should not be used directly, it's essential a object to encapsulate
86
50
the logic in InterRepository.fetch().
88
def __init__(self, to_repository, from_repository, last_revision=None, pb=None):
90
self.failed_revisions = []
92
if to_repository.control_files._transport.base == from_repository.control_files._transport.base:
93
# check that last_revision is in 'from' and then return a no-operation.
94
if last_revision not in (None, NULL_REVISION):
95
from_repository.get_revision(last_revision)
53
def __init__(self, to_repository, from_repository, last_revision=None,
54
find_ghosts=True, fetch_spec=None):
55
"""Create a repo fetcher.
57
:param last_revision: If set, try to limit to the data this revision
59
:param fetch_spec: A SearchResult specifying which revisions to fetch.
60
If set, this overrides last_revision.
61
:param find_ghosts: If True search the entire history for ghosts.
63
# repository.fetch has the responsibility for short-circuiting
64
# attempts to copy between a repository and itself.
97
65
self.to_repository = to_repository
98
66
self.from_repository = from_repository
67
self.sink = to_repository._get_sink()
99
68
# must not mutate self._last_revision as its potentially a shared instance
100
69
self._last_revision = last_revision
102
self.pb = bzrlib.ui.ui_factory.nested_progress_bar()
103
self.nested_pb = self.pb
106
self.nested_pb = None
70
self._fetch_spec = fetch_spec
71
self.find_ghosts = find_ghosts
107
72
self.from_repository.lock_read()
73
mutter("Using fetch logic to copy between %s(%s) and %s(%s)",
74
self.from_repository, self.from_repository._format,
75
self.to_repository, self.to_repository._format)
109
self.to_repository.lock_write()
113
if self.nested_pb is not None:
114
self.nested_pb.finished()
115
self.to_repository.unlock()
117
79
self.from_repository.unlock()
119
81
def __fetch(self):
120
82
"""Primary worker function.
122
This initialises all the needed variables, and then fetches the
84
This initialises all the needed variables, and then fetches the
123
85
requested revisions, finally clearing the progress bar.
125
self.to_weaves = self.to_repository.weave_store
126
self.to_control = self.to_repository.control_weaves
127
self.from_weaves = self.from_repository.weave_store
128
self.from_control = self.from_repository.control_weaves
87
# Roughly this is what we're aiming for fetch to become:
89
# missing = self.sink.insert_stream(self.source.get_stream(search))
91
# missing = self.sink.insert_stream(self.source.get_items(missing))
129
93
self.count_total = 0
130
94
self.file_ids_names = {}
131
pp = ProgressPhase('Fetch phase', 4, self.pb)
134
revs = self._revids_to_fetch()
138
self._fetch_weave_texts(revs)
140
self._fetch_inventory_weave(revs)
142
self._fetch_revision_texts(revs)
143
self.count_copied += len(revs)
95
pb = ui.ui_factory.nested_progress_bar()
96
pb.show_pct = pb.show_count = False
98
pb.update(gettext("Finding revisions"), 0, 2)
99
search_result = self._revids_to_fetch()
100
mutter('fetching: %s', search_result)
101
if search_result.is_empty():
103
pb.update(gettext("Fetching revisions"), 1, 2)
104
self._fetch_everything_for_search(search_result)
108
def _fetch_everything_for_search(self, search):
109
"""Fetch all data for the given set of revisions."""
110
# The first phase is "file". We pass the progress bar for it directly
111
# into item_keys_introduced_by, which has more information about how
112
# that phase is progressing than we do. Progress updates for the other
113
# phases are taken care of in this function.
114
# XXX: there should be a clear owner of the progress reporting. Perhaps
115
# item_keys_introduced_by should have a richer API than it does at the
116
# moment, so that it can feed the progress information back to this
118
if (self.from_repository._format.rich_root_data and
119
not self.to_repository._format.rich_root_data):
120
raise errors.IncompatibleRepositories(
121
self.from_repository, self.to_repository,
122
"different rich-root support")
123
pb = ui.ui_factory.nested_progress_bar()
125
pb.update("Get stream source")
126
source = self.from_repository._get_source(
127
self.to_repository._format)
128
stream = source.get_stream(search)
129
from_format = self.from_repository._format
130
pb.update("Inserting stream")
131
resume_tokens, missing_keys = self.sink.insert_stream(
132
stream, from_format, [])
134
pb.update("Missing keys")
135
stream = source.get_stream_for_missing_keys(missing_keys)
136
pb.update("Inserting missing keys")
137
resume_tokens, missing_keys = self.sink.insert_stream(
138
stream, from_format, resume_tokens)
140
raise AssertionError(
141
"second push failed to complete a fetch %r." % (
144
raise AssertionError(
145
"second push failed to commit the fetch %r." % (
147
pb.update("Finishing stream")
147
152
def _revids_to_fetch(self):
148
mutter('fetch up to rev {%s}', self._last_revision)
149
if self._last_revision is NULL_REVISION:
153
"""Determines the exact revisions needed from self.from_repository to
154
install self._last_revision in self.to_repository.
156
:returns: A SearchResult of some sort. (Possibly a
157
PendingAncestryResult, EmptySearchResult, etc.)
159
if self._fetch_spec is not None:
160
# The fetch spec is already a concrete search result.
161
return self._fetch_spec
162
elif self._last_revision == NULL_REVISION:
163
# fetch_spec is None + last_revision is null => empty fetch.
150
164
# explicit limit of no revisions needed
152
if (self._last_revision is not None and
153
self.to_repository.has_revision(self._last_revision)):
157
return self.to_repository.missing_revision_ids(self.from_repository,
159
except errors.NoSuchRevision:
160
raise InstallFailed([self._last_revision])
162
def _fetch_weave_texts(self, revs):
163
texts_pb = bzrlib.ui.ui_factory.nested_progress_bar()
165
# fileids_altered_by_revision_ids requires reading the inventory
166
# weave, we will need to read the inventory weave again when
167
# all this is done, so enable caching for that specific weave
168
inv_w = self.from_repository.get_inventory_weave()
170
file_ids = self.from_repository.fileids_altered_by_revision_ids(revs)
172
num_file_ids = len(file_ids)
173
for file_id, required_versions in file_ids.items():
174
texts_pb.update("fetch texts", count, num_file_ids)
176
to_weave = self.to_weaves.get_weave_or_empty(file_id,
177
self.to_repository.get_transaction())
178
from_weave = self.from_weaves.get_weave(file_id,
179
self.from_repository.get_transaction())
180
# we fetch all the texts, because texts do
181
# not reference anything, and its cheap enough
182
to_weave.join(from_weave, version_ids=required_versions)
183
# we don't need *all* of this data anymore, but we dont know
184
# what we do. This cache clearing will result in a new read
185
# of the knit data when we do the checkout, but probably we
186
# want to emit the needed data on the fly rather than at the
188
# the from weave should know not to cache data being joined,
189
# but its ok to ask it to clear.
190
from_weave.clear_cache()
191
to_weave.clear_cache()
195
def _fetch_inventory_weave(self, revs):
196
pb = bzrlib.ui.ui_factory.nested_progress_bar()
198
pb.update("fetch inventory", 0, 2)
199
to_weave = self.to_control.get_weave('inventory',
200
self.to_repository.get_transaction())
202
child_pb = bzrlib.ui.ui_factory.nested_progress_bar()
204
# just merge, this is optimisable and its means we don't
205
# copy unreferenced data such as not-needed inventories.
206
pb.update("fetch inventory", 1, 3)
207
from_weave = self.from_repository.get_inventory_weave()
208
pb.update("fetch inventory", 2, 3)
209
# we fetch only the referenced inventories because we do not
210
# know for unselected inventories whether all their required
211
# texts are present in the other repository - it could be
213
to_weave.join(from_weave, pb=child_pb, msg='merge inventory',
215
from_weave.clear_cache()
222
class GenericRepoFetcher(RepoFetcher):
223
"""This is a generic repo to repo fetcher.
225
This makes minimal assumptions about repo layout and contents.
226
It triggers a reconciliation after fetching to ensure integrity.
229
def _fetch_revision_texts(self, revs):
230
"""Fetch revision object texts"""
231
rev_pb = bzrlib.ui.ui_factory.nested_progress_bar()
233
to_txn = self.to_transaction = self.to_repository.get_transaction()
236
to_store = self.to_repository._revision_store
238
pb = bzrlib.ui.ui_factory.nested_progress_bar()
240
pb.update('copying revisions', count, total)
242
sig_text = self.from_repository.get_signature_text(rev)
243
to_store.add_revision_signature_text(rev, sig_text, to_txn)
244
except errors.NoSuchRevision:
247
to_store.add_revision(self.from_repository.get_revision(rev),
252
# fixup inventory if needed:
253
# this is expensive because we have no inverse index to current ghosts.
254
# but on local disk its a few seconds and sftp push is already insane.
256
# FIXME: repository should inform if this is needed.
257
self.to_repository.reconcile()
262
class KnitRepoFetcher(RepoFetcher):
263
"""This is a knit format repository specific fetcher.
265
This differs from the GenericRepoFetcher by not doing a
266
reconciliation after copying, and using knit joining to
270
def _fetch_revision_texts(self, revs):
271
# may need to be a InterRevisionStore call here.
272
from_transaction = self.from_repository.get_transaction()
273
to_transaction = self.to_repository.get_transaction()
274
to_sf = self.to_repository._revision_store.get_signature_file(
276
from_sf = self.from_repository._revision_store.get_signature_file(
278
to_sf.join(from_sf, version_ids=revs, ignore_missing=True)
279
to_rf = self.to_repository._revision_store.get_revision_file(
281
from_rf = self.from_repository._revision_store.get_revision_file(
283
to_rf.join(from_rf, version_ids=revs)
165
return vf_search.EmptySearchResult()
166
elif self._last_revision is not None:
167
return vf_search.NotInOtherForRevs(self.to_repository,
168
self.from_repository, [self._last_revision],
169
find_ghosts=self.find_ghosts).execute()
170
else: # self._last_revision is None:
171
return vf_search.EverythingNotInOther(self.to_repository,
172
self.from_repository,
173
find_ghosts=self.find_ghosts).execute()
286
176
class Inter1and2Helper(object):
287
177
"""Helper for operations that convert data from model 1 and 2
289
179
This is for use by fetchers and converters.
292
def __init__(self, source, target):
182
# This is a class variable so that the test suite can override it.
183
known_graph_threshold = 100
185
def __init__(self, source):
295
188
:param source: The repository data comes from
296
:param target: The repository data goes to
298
190
self.source = source
301
192
def iter_rev_trees(self, revs):
302
193
"""Iterate through RevisionTrees efficiently.
316
209
revs = revs[100:]
211
def _find_root_ids(self, revs, parent_map, graph):
213
for tree in self.iter_rev_trees(revs):
214
root_id = tree.get_root_id()
215
revision_id = tree.get_file_revision(root_id, u"")
216
revision_root[revision_id] = root_id
217
# Find out which parents we don't already know root ids for
219
for revision_parents in parent_map.itervalues():
220
parents.update(revision_parents)
221
parents.difference_update(revision_root.keys() + [NULL_REVISION])
222
# Limit to revisions present in the versionedfile
223
parents = graph.get_parent_map(parents).keys()
224
for tree in self.iter_rev_trees(parents):
225
root_id = tree.get_root_id()
226
revision_root[tree.get_revision_id()] = root_id
318
229
def generate_root_texts(self, revs):
319
230
"""Generate VersionedFiles for all root ids.
321
232
:param revs: the revisions to include
323
inventory_weave = self.source.get_inventory_weave()
326
to_store = self.target.weave_store
327
for tree in self.iter_rev_trees(revs):
328
revision_id = tree.inventory.root.revision
329
root_id = tree.inventory.root.file_id
330
parents = inventory_weave.get_parents(revision_id)
331
if root_id not in versionedfile:
332
versionedfile[root_id] = to_store.get_weave_or_empty(root_id,
333
self.target.get_transaction())
334
parent_texts[root_id] = versionedfile[root_id].add_lines(
335
revision_id, parents, [], parent_texts)
337
def regenerate_inventory(self, revs):
338
"""Generate a new inventory versionedfile in target, convertin data.
340
The inventory is retrieved from the source, (deserializing it), and
341
stored in the target (reserializing it in a different format).
342
:param revs: The revisions to include
344
inventory_weave = self.source.get_inventory_weave()
345
for tree in self.iter_rev_trees(revs):
346
parents = inventory_weave.get_parents(tree.get_revision_id())
347
self.target.add_inventory(tree.get_revision_id(), tree.inventory,
351
class Model1toKnit2Fetcher(GenericRepoFetcher):
352
"""Fetch from a Model1 repository into a Knit2 repository
354
def __init__(self, to_repository, from_repository, last_revision=None,
356
self.helper = Inter1and2Helper(from_repository, to_repository)
357
GenericRepoFetcher.__init__(self, to_repository, from_repository,
360
def _fetch_weave_texts(self, revs):
361
GenericRepoFetcher._fetch_weave_texts(self, revs)
362
# Now generate a weave for the tree root
363
self.helper.generate_root_texts(revs)
365
def _fetch_inventory_weave(self, revs):
366
self.helper.regenerate_inventory(revs)
369
class Knit1to2Fetcher(KnitRepoFetcher):
370
"""Fetch from a Knit1 repository into a Knit2 repository"""
372
def __init__(self, to_repository, from_repository, last_revision=None,
374
self.helper = Inter1and2Helper(from_repository, to_repository)
375
KnitRepoFetcher.__init__(self, to_repository, from_repository,
378
def _fetch_weave_texts(self, revs):
379
KnitRepoFetcher._fetch_weave_texts(self, revs)
380
# Now generate a weave for the tree root
381
self.helper.generate_root_texts(revs)
383
def _fetch_inventory_weave(self, revs):
384
self.helper.regenerate_inventory(revs)
387
class Fetcher(object):
388
"""Backwards compatibility glue for branch.fetch()."""
390
@deprecated_method(zero_eight)
391
def __init__(self, to_branch, from_branch, last_revision=None, pb=None):
392
"""Please see branch.fetch()."""
393
to_branch.fetch(from_branch, last_revision, pb)
234
graph = self.source.get_graph()
235
parent_map = graph.get_parent_map(revs)
236
rev_order = tsort.topo_sort(parent_map)
237
rev_id_to_root_id = self._find_root_ids(revs, parent_map, graph)
238
root_id_order = [(rev_id_to_root_id[rev_id], rev_id) for rev_id in
240
# Guaranteed stable, this groups all the file id operations together
241
# retaining topological order within the revisions of a file id.
242
# File id splits and joins would invalidate this, but they don't exist
243
# yet, and are unlikely to in non-rich-root environments anyway.
244
root_id_order.sort(key=operator.itemgetter(0))
245
# Create a record stream containing the roots to create.
246
if len(revs) > self.known_graph_threshold:
247
graph = self.source.get_known_graph_ancestry(revs)
248
new_roots_stream = _new_root_data_stream(
249
root_id_order, rev_id_to_root_id, parent_map, self.source, graph)
250
return [('texts', new_roots_stream)]
253
def _new_root_data_stream(
254
root_keys_to_create, rev_id_to_root_id_map, parent_map, repo, graph=None):
255
"""Generate a texts substream of synthesised root entries.
257
Used in fetches that do rich-root upgrades.
259
:param root_keys_to_create: iterable of (root_id, rev_id) pairs describing
260
the root entries to create.
261
:param rev_id_to_root_id_map: dict of known rev_id -> root_id mappings for
262
calculating the parents. If a parent rev_id is not found here then it
263
will be recalculated.
264
:param parent_map: a parent map for all the revisions in
266
:param graph: a graph to use instead of repo.get_graph().
268
for root_key in root_keys_to_create:
269
root_id, rev_id = root_key
270
parent_keys = _parent_keys_for_root_version(
271
root_id, rev_id, rev_id_to_root_id_map, parent_map, repo, graph)
272
yield versionedfile.FulltextContentFactory(
273
root_key, parent_keys, None, '')
276
def _parent_keys_for_root_version(
277
root_id, rev_id, rev_id_to_root_id_map, parent_map, repo, graph=None):
278
"""Get the parent keys for a given root id.
280
A helper function for _new_root_data_stream.
282
# Include direct parents of the revision, but only if they used the same
283
# root_id and are heads.
284
rev_parents = parent_map[rev_id]
286
for parent_id in rev_parents:
287
if parent_id == NULL_REVISION:
289
if parent_id not in rev_id_to_root_id_map:
290
# We probably didn't read this revision, go spend the extra effort
293
tree = repo.revision_tree(parent_id)
294
except errors.NoSuchRevision:
295
# Ghost, fill out rev_id_to_root_id in case we encounter this
297
# But set parent_root_id to None since we don't really know
298
parent_root_id = None
300
parent_root_id = tree.get_root_id()
301
rev_id_to_root_id_map[parent_id] = None
303
# rev_id_to_root_id_map[parent_id] = parent_root_id
304
# memory consumption maybe?
306
parent_root_id = rev_id_to_root_id_map[parent_id]
307
if root_id == parent_root_id:
308
# With stacking we _might_ want to refer to a non-local revision,
309
# but this code path only applies when we have the full content
310
# available, so ghosts really are ghosts, not just the edge of
312
parent_ids.append(parent_id)
314
# root_id may be in the parent anyway.
316
tree = repo.revision_tree(parent_id)
317
except errors.NoSuchRevision:
318
# ghost, can't refer to it.
322
parent_ids.append(tree.get_file_revision(root_id))
323
except errors.NoSuchId:
326
# Drop non-head parents
328
graph = repo.get_graph()
329
heads = graph.heads(parent_ids)
331
for parent_id in parent_ids:
332
if parent_id in heads and parent_id not in selected_ids:
333
selected_ids.append(parent_id)
334
parent_keys = [(root_id, parent_id) for parent_id in selected_ids]
338
class TargetRepoKinds(object):
339
"""An enum-like set of constants.
341
They are the possible values of FetchSpecFactory.target_repo_kinds.
344
PREEXISTING = 'preexisting'
349
class FetchSpecFactory(object):
350
"""A helper for building the best fetch spec for a sprout call.
352
Factors that go into determining the sort of fetch to perform:
353
* did the caller specify any revision IDs?
354
* did the caller specify a source branch (need to fetch its
355
heads_to_fetch(), usually the tip + tags)
356
* is there an existing target repo (don't need to refetch revs it
358
* target is stacked? (similar to pre-existing target repo: even if
359
the target itself is new don't want to refetch existing revs)
361
:ivar source_branch: the source branch if one specified, else None.
362
:ivar source_branch_stop_revision_id: fetch up to this revision of
363
source_branch, rather than its tip.
364
:ivar source_repo: the source repository if one found, else None.
365
:ivar target_repo: the target repository acquired by sprout.
366
:ivar target_repo_kind: one of the TargetRepoKinds constants.
370
self._explicit_rev_ids = set()
371
self.source_branch = None
372
self.source_branch_stop_revision_id = None
373
self.source_repo = None
374
self.target_repo = None
375
self.target_repo_kind = None
378
def add_revision_ids(self, revision_ids):
379
"""Add revision_ids to the set of revision_ids to be fetched."""
380
self._explicit_rev_ids.update(revision_ids)
382
def make_fetch_spec(self):
383
"""Build a SearchResult or PendingAncestryResult or etc."""
384
if self.target_repo_kind is None or self.source_repo is None:
385
raise AssertionError(
386
'Incomplete FetchSpecFactory: %r' % (self.__dict__,))
387
if len(self._explicit_rev_ids) == 0 and self.source_branch is None:
388
if self.limit is not None:
389
raise NotImplementedError(
390
"limit is only supported with a source branch set")
391
# Caller hasn't specified any revisions or source branch
392
if self.target_repo_kind == TargetRepoKinds.EMPTY:
393
return vf_search.EverythingResult(self.source_repo)
395
# We want everything not already in the target (or target's
397
return vf_search.EverythingNotInOther(
398
self.target_repo, self.source_repo).execute()
399
heads_to_fetch = set(self._explicit_rev_ids)
400
if self.source_branch is not None:
401
must_fetch, if_present_fetch = self.source_branch.heads_to_fetch()
402
if self.source_branch_stop_revision_id is not None:
403
# Replace the tip rev from must_fetch with the stop revision
404
# XXX: this might be wrong if the tip rev is also in the
405
# must_fetch set for other reasons (e.g. it's the tip of
406
# multiple loom threads?), but then it's pretty unclear what it
407
# should mean to specify a stop_revision in that case anyway.
408
must_fetch.discard(self.source_branch.last_revision())
409
must_fetch.add(self.source_branch_stop_revision_id)
410
heads_to_fetch.update(must_fetch)
412
if_present_fetch = set()
413
if self.target_repo_kind == TargetRepoKinds.EMPTY:
414
# PendingAncestryResult does not raise errors if a requested head
415
# is absent. Ideally it would support the
416
# required_ids/if_present_ids distinction, but in practice
417
# heads_to_fetch will almost certainly be present so this doesn't
419
all_heads = heads_to_fetch.union(if_present_fetch)
420
ret = vf_search.PendingAncestryResult(all_heads, self.source_repo)
421
if self.limit is not None:
422
graph = self.source_repo.get_graph()
423
topo_order = list(graph.iter_topo_order(ret.get_keys()))
424
result_set = topo_order[:self.limit]
425
ret = self.source_repo.revision_ids_to_search_result(result_set)
428
return vf_search.NotInOtherForRevs(self.target_repo, self.source_repo,
429
required_ids=heads_to_fetch, if_present_ids=if_present_fetch,
430
limit=self.limit).execute()